Article type
Abstract
Background:
The use of Bayesian statistics for designing randomized clinical trials (RCTs) has increased in the past decade. This has enabled trialists to incorporate historical data when designing trials, make predictions that appropriately integrate multiple sources of uncertainty during a trial, and plan interim analyses for data monitoring. These features make Bayesian approaches a viable alternative to frequentist design. There is, however, significant variation in the application of these methods in existing trials. The unstandardized documentation and reporting of contemporary Bayesian methods may hinder their interpretability, credibility, and wider implementation in randomized controlled trials. Thus, we aim to conduct a systematic review of clinical trials using Bayesian approaches for their primary analysis to describe the current methodological landscape.
Methods:
We searched MEDLINE and Embase between 2019 and 2023 for RCTs that used Bayesian methodology. The Cochrane Central Register of Controlled Trials (CENTRAL) was used to identify trial protocols in the same period. Pairs of reviewers will independently screen studies for inclusion, resolving disagreements by consensus with expert methodologist review. We will exclude non-English studies and publications without obtainable full-texts. We will exclude cluster, n-of-1, and phase 1 trials. We designed a data extraction form to collect relevant methodologic items including: basic trial information, sample size considerations, adaptive design (e.g. decision rules, early stopping), data analysis, Bayesian parameters (e.g. priors), and results reporting. Results will be stratified using the 2023 clinically relevant journals filter, trial phase, clinical area, and other key methodological features.
Results:
We identified 1971 citations, of which 448 were duplicates and 278 were marked ineligible by the Covidence automated RCT screening tool. Of 1245 studies identified for screening, 346 were from clinically useful journals. Of the 327 citations identified from CENTRAL, 253 were sourced from clinicaltrials.gov and 74 were sourced from the International Clinical Trials Registry Platform (ICTRP). We hypothesize that there will be significant variation in methodologic approaches across studies.
Discussion:
Our findings will provide insight into the contemporary landscape of Bayesian approaches to designing modern clinical trials, including current best practices and opportunities for improvement, and inform future guidelines for applying Bayesian methods to trials.
The use of Bayesian statistics for designing randomized clinical trials (RCTs) has increased in the past decade. This has enabled trialists to incorporate historical data when designing trials, make predictions that appropriately integrate multiple sources of uncertainty during a trial, and plan interim analyses for data monitoring. These features make Bayesian approaches a viable alternative to frequentist design. There is, however, significant variation in the application of these methods in existing trials. The unstandardized documentation and reporting of contemporary Bayesian methods may hinder their interpretability, credibility, and wider implementation in randomized controlled trials. Thus, we aim to conduct a systematic review of clinical trials using Bayesian approaches for their primary analysis to describe the current methodological landscape.
Methods:
We searched MEDLINE and Embase between 2019 and 2023 for RCTs that used Bayesian methodology. The Cochrane Central Register of Controlled Trials (CENTRAL) was used to identify trial protocols in the same period. Pairs of reviewers will independently screen studies for inclusion, resolving disagreements by consensus with expert methodologist review. We will exclude non-English studies and publications without obtainable full-texts. We will exclude cluster, n-of-1, and phase 1 trials. We designed a data extraction form to collect relevant methodologic items including: basic trial information, sample size considerations, adaptive design (e.g. decision rules, early stopping), data analysis, Bayesian parameters (e.g. priors), and results reporting. Results will be stratified using the 2023 clinically relevant journals filter, trial phase, clinical area, and other key methodological features.
Results:
We identified 1971 citations, of which 448 were duplicates and 278 were marked ineligible by the Covidence automated RCT screening tool. Of 1245 studies identified for screening, 346 were from clinically useful journals. Of the 327 citations identified from CENTRAL, 253 were sourced from clinicaltrials.gov and 74 were sourced from the International Clinical Trials Registry Platform (ICTRP). We hypothesize that there will be significant variation in methodologic approaches across studies.
Discussion:
Our findings will provide insight into the contemporary landscape of Bayesian approaches to designing modern clinical trials, including current best practices and opportunities for improvement, and inform future guidelines for applying Bayesian methods to trials.